What 1% better gets wrong
By Ayush Mishra
One percent better is a compound-interest cartoon. Automaticity does not rise in a straight line, and a missed day is not a 0.99 shock to your identity.
If you want this next to the Atomic Habits product, Habit AI vs Atoms is the honest comparison.
"1% better every day" is a compound-interest cartoon that escaped a book and colonised LinkedIn. James Clear used it in Atomic Habits as a metaphor for small gains accumulating. Dave Brailsford's "aggregation of marginal gains" was a cycling-programme slogan, not a law of behaviour. Neither is how automaticity actually grows. Phillippa Lally, van Jaarsveld, Potts, and Wardle (2010) fitted an asymptotic curve: automaticity rose with repetitions of a defined act in a stable context, faster at first for many people, then levelling, with a median around 66 days and a range from 18 to 254. That is not 1.01 to the power of n.
The cartoon harms in three specific ways. It implies a straight line. It implies that a missed day is a 0.99 shock, which then becomes a story about ruin. It implies that any small thing counts as 1%, which licenses busywork (reorganising the app, buying a new notebook) in place of the behaviour.
Curves, not interest rates
Lally's group also reported that a single missed day did not, on average, meaningfully derail the automaticity trajectory. That finding is the opposite of compound-interest folklore, where a zero day is a crash. Missing one day: what the evidence says is the full note. What a habit streak actually measures is why a chain is the wrong picture of that curve.
Wendy Wood's context-cued automaticity is not a percentage. It is an association that strengthens with co-occurrence of setting and act. If the setting vanishes (holiday, new baby, new commute), the association does not politely lose 1%. It fails to fire, then can be rebuilt in the new context. Habit discontinuity is that break. Marginal gains do not describe it.
Gollwitzer's if-then plans do not compound. They either get installed on a cue or they do not. You can have a beautiful 1% journal and no if-then, which is how people stay at 0% on the only metric that matters: did the action occur after the cue.
Charles Duhigg popularised a loop you inspect, not a rate you compound. Fogg's tiny habits are a shrink, not a 1% levy on your soul. Clear's four laws are still usable if you drop the cartoon. Atomic Habits in practice is the workshop without the interest rate.
What people do instead of the behaviour
They collect 1% tasks: a new highlighter, a new icon, a new "system" document. That is nearby, effortful, and easy to count. It is not the walk. Goodhart again: the measure (did I improve something) ate the goal (did I do the thing).
They treat rest days as moral debt. Lally says the learning usually survives a miss. Marlatt's abstinence violation effect says the story after a miss is what takes the week. The 1% cartoon feeds the story: you are now behind the compound curve, so why bother. Relapse is data is the better story.
They average incomparable acts. A 1% better inbox-zero and a 1% better sleep wind-down are not addable. There is no shared unit. Choosing one metric per habit exists because addition across rows is how dashboards lie.
Brailsford's cycling programme, as journalists told it, was about many small technical improvements on a team with coaches, equipment, and a season. You do not have a wind-tunnel in your kitchen. Stealing the slogan without the lab is how a stretch becomes "marginal gains" and then becomes nothing.
| Claim | What the science says | What to do instead |
|---|---|---|
| 1% better daily compounds like money | Automaticity is an asymptotic curve (Lally 2010) | Count repetitions in a stable context |
| A miss multiplies you by 0.99 and ruins the year | A miss did not, on average, wreck the curve | Do the next tiny instance after the same cue |
| Any small improvement counts | Cues and defined acts count | One sentence, one metric, no busywork |
| Identity is the compound interest of votes | Votes are repetitions, not captions | Open the book; skip the caption |
How long does it actually take to form a habit is the 66-day evidence, including the spread. If you need a number, use a range and a median, not 1.01^365 ≈ 37.8, which is a maths party trick about a metaphor. The trick also assumes every day is a gain of equal size. A two-minute walk and a heroic Saturday session are not equal deposits in a bank. Wood's association cares that the cue and the act co-occurred. It does not pay interest on intensity theatre.
Consistency percentage over 28 days is a better dashboard than a fantasy interest rate. Streaks versus consistency percentage and habit analytics are the pictures that survive contact with a Thursday.
Habit AI is a useful example of dropping the cartoon in product form: partial credit and consistency over a window, rather than a compound-interest identity. The behaviour is still the walk. The metaphor can stay on LinkedIn.
If 1% better helped you start, keep the starting. Drop the physics. You do not need to be 37 times yourself by December. You need the same cue, a small act, and enough repetitions that Wood would recognise the setting as loaded. That is slower to brag about. It is what the curves actually show.
Read next
How long does it actually take to form a habit is the real curve. Atomic Habits in practice is the rest of the field guide with the cartoon stripped. For a log that treats a miss as a miss, see how Habit AI works.
Share